U.S. patent application number 16/867623 was filed with the patent office on 2020-11-26 for determination of leak during cpap treatment.
This patent application is currently assigned to ResMed Pty Ltd. The applicant listed for this patent is ResMed Pty Ltd. Invention is credited to David John BASSIN.
Application Number | 20200368467 16/867623 |
Document ID | / |
Family ID | 1000005004369 |
Filed Date | 2020-11-26 |
![](/patent/app/20200368467/US20200368467A1-20201126-D00001.png)
![](/patent/app/20200368467/US20200368467A1-20201126-D00002.png)
![](/patent/app/20200368467/US20200368467A1-20201126-M00001.png)
![](/patent/app/20200368467/US20200368467A1-20201126-M00002.png)
United States Patent
Application |
20200368467 |
Kind Code |
A1 |
BASSIN; David John |
November 26, 2020 |
DETERMINATION OF LEAK DURING CPAP TREATMENT
Abstract
A respiratory treatment apparatus and method in which a leak is
determined by using an averaging window. The window starts at the
present time and extends back in time to a point determined
according to a current one of progressively detected phase measures
of a first respiratory cycle and a corresponding phase measure
attributable to a preceding second respiratory cycle. In another
aspect, a jamming index indicates whether the leak is rapidly
changing. To the extent that jamming is high, the leak estimate
used progressively changes from that using sliding breath-window
averaging to a more robust and faster responding low-pass filter
method, and adjustment of ventilatory support based on measures
employing estimated respiratory flow is slowed down or stopped.
Inventors: |
BASSIN; David John; (Coogee,
AU) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
ResMed Pty Ltd |
Bella Vista |
|
AU |
|
|
Assignee: |
ResMed Pty Ltd,
Bella Vista
AU
|
Family ID: |
1000005004369 |
Appl. No.: |
16/867623 |
Filed: |
May 6, 2020 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15203427 |
Jul 6, 2016 |
10675424 |
|
|
16867623 |
|
|
|
|
13874668 |
May 1, 2013 |
9408989 |
|
|
15203427 |
|
|
|
|
13542331 |
Jul 5, 2012 |
8459260 |
|
|
13874668 |
|
|
|
|
12438758 |
Feb 25, 2009 |
8256418 |
|
|
PCT/AU2007/001237 |
Aug 30, 2007 |
|
|
|
13542331 |
|
|
|
|
60823934 |
Aug 30, 2006 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
A61M 16/024 20170801;
A61M 2016/0036 20130101; A61M 16/06 20130101; A61M 2205/50
20130101; A61M 16/0057 20130101; A61M 2205/16 20130101; A61M
2205/15 20130101; A61M 2016/0021 20130101 |
International
Class: |
A61M 16/00 20060101
A61M016/00; A61M 16/06 20060101 A61M016/06 |
Claims
1. A method of determining a mask leak conductance in a patient
respiratory device, comprising: calculating, with filtered measures
of flow and pressure, a first estimate of leak conductance which
responds relatively quickly to changes in leak conductance,
calculating, with the filtered measures of flow and pressure, a
second estimate of leak conductance which responds more slowly to
changes in leak conductance, and determining the mask leak
conductance as a function of the first and second estimates of leak
conductance.
2. The method of claim 1, further comprising determining pressure
in a mask of the patient respiratory device.
3. The method of claim 2, further comprising determining an
estimate of non-deliberate mask leak as a function of the mask leak
conductance and the mask pressure.
4. The method of claim 3, wherein the non-deliberate mask leak is
calculated as: Q.sub.leak=G.sub.j {square root over ( )}P.sub.mask
wherein: Q.sub.leak is non-deliberate mask leak, G.sub.j is mask
leak conductance, and P.sub.mask is mask pressure.
5. The method of claim 1, further comprising determining an
uncompensated leak index representative of an extent to which mask
leak is uncompensated.
6. The method of claim 5, wherein the mask leak conductance is
calculated as a function of the uncompensated leak index and the
first and second estimates of leak conductance.
7. The method of claim 6, wherein the mask leak conductance is
calculated as: G.sub.j=JG.sub.1+(1-J)G.sub.2 wherein: G.sub.j is an
estimate of mask leak conductance, G.sub.1 is the first estimate of
leak conductance, G.sub.2 is the second estimate of leak
conductance, and J is a jamming index of the uncompensated mask
leak.
8. The method of claim 7, wherein the jamming index is between zero
and one.
9. The method of claim 1 wherein the respiratory device comprises a
controller for controlling a ventilator, wherein the controller
applies the mask leak conductance to adjust an output of the
ventilator.
10. The method of claim 9 wherein the output comprises a pressure
support level.
11. The method of claim 1 wherein the filtered measures of flow and
pressure are derived by low pass filtering.
12. The method of claim 11 wherein the filtered measures of flow
and pressure are further derived by averaging over a window.
13. Respiratory apparatus for determining a mask leak conductance
comprising: sensors configured to sense flow and pressure
associated with a mask of the respiratory apparatus; and a
controller coupled with the sensors, the controller configured to:
calculate, with filtered measures of the flow and pressure, a first
estimate of leak conductance which responds relatively quickly to
changes in leak conductance, calculate, with the filtered measures
of the flow and pressure, a second estimate of leak conductance
which responds more slowly to changes in leak conductance, and
determine the mask leak conductance as a function of the first and
second estimates of leak conductance.
14. The respiratory apparatus of claim 13, wherein the controller
is further configured to determine an estimate of non-deliberate
mask leak as a function of the mask leak conductance and the mask
pressure.
15. The respiratory apparatus of claim 14, wherein the controller
is configured to calculate the non-deliberate mask leak as:
Q.sub.leak=G.sub.j {square root over ( )}P.sub.mask wherein:
Q.sub.leak is non-deliberate mask leak, G.sub.j is mask leak
conductance, and P.sub.mask is mask pressure.
16. The respiratory apparatus of claim 13, wherein the controller
is further configured to determine an uncompensated leak index
representative of an extent to which mask leak is
uncompensated.
17. The respiratory apparatus of claim 16, wherein the controller
is configured to calculate that the mask leak conductance as a
function of the uncompensated leak index and the first and second
estimates of leak conductance.
18. The respiratory apparatus of claim 17, wherein the controller
is configured to calculate the mask leak conductance as:
G.sub.j=JG.sub.1+(1-J)G.sub.2 wherein: G.sub.j is an estimate of
mask leak conductance, G.sub.1 is the first estimate of leak
conductance, G.sub.2 is the second estimate of leak conductance,
and J is a jamming index of the uncompensated mask leak.
19. The respiratory apparatus of claim 18, wherein the jamming
index is between zero and one.
20. The respiratory apparatus of claim 13 wherein the respiratory
apparatus comprises a ventilator, and wherein the controller is
configured to apply the mask leak conductance to adjust an output
of the ventilator.
21. The respiratory apparatus of claim 20 wherein the output
comprises a pressure support level.
22. The respiratory apparatus of claim 13 wherein the controller is
configured to derive the filtered measures of flow and pressure by
low pass filtering.
23. The respiratory apparatus of claim 22 wherein the controller is
configured to further derive the filtered measures of flow and
pressure by averaging over a window.
24. The respiratory apparatus of claim 13 wherein the sensors
comprise a pressure sensor and pneumotachograph.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] The present application is a continuation of U.S. patent
application Ser. No. 15/203,427 filed Jul. 6, 2016, which is a
divisional of U.S. patent application Ser. No. 13/874,668 filed May
1, 2013, now U.S. Pat. No. 9,408,989, which is a continuation of
U.S. patent application Ser. No. 13/542,331 filed Jul. 5, 2012, now
U.S. Pat. No. 8,459,260, which is a continuation of U.S. patent
application Ser. No. 12/438,758 filed Feb. 25, 2009, now U.S. Pat.
No. 8,256,418 which is a national phase entry under 35 U.S.C.
.sctn. 371 of International Application No. PCT/AU 2007/001237
filed Aug. 30, 2007, which claims the benefit of the filing date of
U.S. Provisional Application No. 60/823,934 filed Aug. 30, 2006,
the disclosures of all of which are incorporated herein by
reference.
FIELD OF THE INVENTION
[0002] This invention relates to treatment of apneas and other
respiratory disorders. In particular it relates to methods and
apparatus for the determination of leakage airflow and true
respiratory airflow, during the mechanical application of positive
airway pressure.
BACKGROUND OF THE INVENTION
[0003] For the treatment of apneas and other respiratory disorders,
breathable gas is supplied from a mechanical respirator or
ventilator, for example via a mask, at a pressure which may be
higher during inspiration and lower during expiration. (In this
specification any reference to a "mask" is to be understood as
including all forms of devices for passing breathable gas to a
person's airway, including nose masks, nose and mouth masks, nasal
prongs/pillows and endrotracheal or tracheostomy tubes. The term
"ventilator" is used to describe any device that does part of the
work of breathing.) Typically one measures the subject's
respiratory airflow during mechanical ventilation to assess
adequacy of treatment, or to control the operation of the
ventilator.
[0004] Respiratory airflow is commonly measured with a
pneumotachograph placed in the gas delivery path between the mask
and the pressure source. Leaks between the mask and the subject are
unavoidable. The pneumotachograph measures the sum of the
respiratory airflow plus the flow through the leak plus flow
through the vent (also called "deliberate leak"). If the
instantaneous flow through the leak is known, the respiratory
airflow can be calculated by subtracting the flow through the leak
and the flow through the vent from the flow at the pneumotach.
Typically the flow through the vent is a known function of pressure
at the vent, and given that the pressure at the vent can be
estimated with reasonable accuracy, the flow through the vent can
then be straightforwardly calculated. On the other hand, if the
vent characteristics are suitable for the leak model employed, the
vent flow and non-deliberate leak can be lumped together and
estimated as a single quantity. The direct estimation of vent flow
using pressure at the vent will be assumed hereinafter, and
subtraction of this vent flow from total ventilator outflow will be
assumed to have occurred when not mentioned explicitly.
[0005] Some methods to correct for the flow through the leak assume
(i) that the leak is substantially constant, and (ii) that over a
sufficiently long time, inspiratory and expiratory respiratory
airflow will cancel. If these assumptions are met, the average flow
through the pneumotach over a sufficiently long period will equal
the magnitude of the leak, and the true respiratory airflow may
then be calculated as described.
[0006] The known method is only correct if the pressure at the mask
is constant. If, on the other hand, the mask pressure varies with
time (for example, in the case of a ventilator), assumption (i)
above will be invalid, and the calculated respiratory airflow will
therefore be incorrect. This is shown markedly in FIGS. 1a-1f.
[0007] FIG. 1a shows a trace of measured mask pressure in bi-level
CPAP (Continuous Positive Airway Pressure) treatment between about
4 cm H2O on expiration and 12 cm H2O on inspiration. FIG. 1b shows
a trace of true respiratory airflow in synchronism with the mask
pressures. At time=21 seconds a mask leak occurs, resulting in a
leakage flow from the leak that is a function of the treatment
pressure, as shown in FIG. 1c. The measured mask flow shown in FIG.
1d now includes an offset due to the leak flow. The prior art
method then determines the calculated leak flow over a number of
breaths, as shown in FIG. 1e. The resulting calculated respiratory
flow, as the measured flow minus the calculating leak flow is shown
in FIG. 1f, having returned to the correct mean value, however is
incorrectly scaled in magnitude, giving a false indication of peak
positive and negative airflow.
[0008] Another prior art arrangement is disclosed in European
Publication No. 0 714 670 A2, including a calculation of a
pressure-dependent leak component. The methodology relies on
knowing precisely the occurrence of the start of an inspiratory
event and the start of the next inspiratory event. In other words,
the leak calculation is formed as an average over a known breath
and applied to a subsequent breath.
[0009] This method cannot be used if the moment of start and end of
the previous breath are unknown. In general, it can be difficult to
accurately calculate the time of start of a breath. This is
particularly the case immediately following a sudden change in the
leak.
[0010] Furthermore, the method will not work in the case of a
subject who is making no respiratory efforts, and is momentarily
not being ventilated at all, for example during an apnea, because
for the duration of the apnea there is no start or end of breath
over which to make a calculation.
[0011] In U.S. Pat. No. 6,162,129 (Berthon-Jones) the leak is
determined by first estimating the conductance of the leak path
from the long term orifice flow:
1 R L = Q P , ##EQU00001##
where G.sub.L=1/R.sub.L is conductance (L denotes leak), Q is
instantaneous flow, p is instantaneous pressure and < >
denotes a long term average calculated for example by low pass
filtering with an IIF or other filter having a long time constant.
Note that the word "average" as used herein contains the general
sense inclusive of the result of a low pass filtering step, and is
not limited to an arithmetic mean or other standard average such as
the RMS average. The instantaneous leak flow, based on the model of
the flow through an orifice is then
Q L = 1 R L p ##EQU00002##
[0012] Note that the instantaneous respiratory airflow is then
Q.sub.R=Q-Q.sub.L.
[0013] Berthon-Jones attempts to deal with sudden changes in
instantaneous leak flow by dynamically adjusting the filter's time
constant using fuzzy logic, lengthening the time constant if it is
certain that the leak is steady, reducing the time constant if it
is certain that the leak has suddenly changed, and using
intermediately longer or shorter time constants if it is
intermediately certain that the leak is steady.
[0014] Berthon-Jones also develops a jamming index by fuzzy logic
to deal with the case of a large and sudden increase in the
conductance of the leak, in which case the calculated respiratory
airflow will be incorrect. In particular during apparent
inspiration, the calculated respiratory airflow will be large
positive for a time that is large compared with the expected
duration of a normal inspiration. Conversely, if there is a sudden
decrease in conductance of the leak, then during apparent
expiration the calculated respiratory airflow will be large
negative for a time that is large compared with the duration of
normal expiration.
[0015] Therefore, the jamming index, i.e. an index of the degree of
certainty that the leak has suddenly changed, is derived, such that
the longer the airflow has been away from zero, and by a larger
amount, the larger the index. The explicit calculation of the
jamming index by fuzzy logic is described in the '129 patent, which
is incorporated herein by reference.
[0016] The time constant for the low pass filters is then adjusted
to vary inversely with the jamming index. In operation, if there is
a sudden and large change in the leak, the index will be large, and
the time constant for the calculation of the conductance of the
leak will be small, allowing rapid convergence on the new value of
the leakage conductance. Conversely, if the leak is steady for a
long time, the index will be small, and the time constant for
calculation of the leakage conductance will be large; enabling
accurate calculation of the instantaneous respiratory airflow. In
the spectrum of intermediate situations, where the calculated
instantaneous respiratory airflow is larger and for longer periods,
the index will be progressively larger, and the time constant for
the calculation of the leak will progressively reduce. For example,
at a moment in time where it is uncertain whether the leak is in
fact constant, and the subject merely commenced a large sigh, or
whether in fact there has been a sudden increase in the leak, the
index will be of an intermediate value, and the time constant for
calculation of the impedance of the leak will also be of an
intermediate value.
BRIEF DESCRIPTION OF THE DRAWINGS
[0017] FIG. 1a shows a trace of measured marked pressure in
bi-level CPAP treatment.
[0018] FIG. 1b shows a trace of true respiratory airflow in
synchronism with mark pressures.
[0019] FIG. 1c shows a trace of leakage flow that is a function of
treatment pressure.
[0020] FIG. 1d shows a trace that illustrates an offset due to leak
flow.
[0021] FIG. 1e shows calculated leak flows over a number of
breaths.
[0022] FIG. 1f shows calculated respiratory flow.
BRIEF DESCRIPTION OF THE INVENTION
[0023] This invention rapidly determines the instantaneous leak in
a CPAP system without detailed modeling the source of the leak and
without having to determine the precise phase in a breathing cycle
at which the leak occurs. It relies instead on the use of timers to
define the breathing cycle and a calculation to assure that the
instantaneous flow is compared to the flow over a time period long
enough to include an entire breath. It does this by looking
backward to include an entire phase cycle. This avoids having to
take long term averages over multiple breaths, or to have a model
that recognized the beginning and end of a breath.
[0024] Sudden changes in a leak are recognized and the degree to
which leak is rapidly changing is expressed as a jamming index
value, which is then used as a parameter to adjust the
contributions of the components of which the leak estimate is made
up, and, in the case of a servoventilator, to temporarily slow down
or suspend the adjustment of the servoventilator controller output
parameter, typically pressure support level.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
[0025] The present invention is motivated by the desire to produce
a continuously updated estimate of the leak model parameter which
is very stable when the actual leak parameter is stable, but which
degrades progressively and gracefully to a less stable, less
accurate but faster-responding estimate when the actual leak
parameter is changing rapidly. The leak model parameter in question
is typically an orifice constant (equivalently a leak conductance),
but need not be.
[0026] A continuously updated estimate of leak conductance (in
particular, continuously updated during each breath) may be
calculated by performing some kind of low-pass filtering operation,
such as a 1st-order low pass filter or a moving average filter,
typically with a fixed window length, on the non-vent flow (equal
to the sum of the respiratory flow and the instantaneous leak flow)
and on the square root of the mask pressure, producing a leak
conductance estimate G.sub.1, as in Berthon-Jones. This method has
the advantage over some other methods that it is independent of the
determination of breath phase (the position within the current
breath). Thus sudden changes in leak can occur which cause
respiratory flow estimates and hence breath phase estimates to be
grossly in error, yet updating of the leak parameter estimates
continues. A disadvantage of such breath-independent estimates is
that the estimates are not stable within a breath unless particular
fortuitous events occur; for example, by coincidence, the duration
of a window averaging filter includes exactly N breaths, where N is
an integer. A particular case of this instability is illustrated by
considering the situation when 1st order low-pass-filter estimates
of mask flow and mask pressure are used. For simplicity, assume
that a mask pressure is constant, and that true leak conductance is
zero. Then the leak flow estimate is just a 1st-order
low-pass-filtered version of respiratory flow. This estimate rises
whenever respiratory flow is above the leak flow estimate, and
falls when respiratory flow is below the leak flow estimate. In
particular, with reasonable filter time constants, the leak flow
estimate rises during most of inspiration and falls during most of
inspiration, rising slowly during the expiratory pause, and under
normal circumstances crucially being below zero during the
expiratory pause. Since true respiratory flow is zero during the
expiratory pause, estimated respiratory flow, being the difference
between mask flow (by assumption equal to respiratory flow) and
estimated leak flow, is positive during the expiratory pause, say
equal to Q.sub.eps. If a ventilator is designed to trigger into
inspiration when the estimated respiratory flow exceeds some true
respiratory flow threshold Q.sub.inso_thresh, a ventilator which
uses this kind of leak estimate, in order to trigger at the desired
true respiratory flow, must set its trigger threshold to a higher
value Q.sub.insp_thresh+Q.sub.eps. Unfortunately Q.sub.eps is a
function of the respiratory flow shape, the respiratory rate, and
the low-pass-filter time constant, very difficult if not impossible
to determine in real time. Hence triggering actually occurs at a
variable threshold, and in the worst case auto-triggering
(triggering at a true respiratory flow of zero) may occur. It
should be noted that the effect of the non-constant estimate of
leak parameter in producing a distorted respiratory flow exists
throughout the breath and whether there is an identifiable
expiratory pause or not, with potential adverse effects on cycling
(expiratory triggering) as well an on (inspiratory) triggering, as
well as other algorithms which operate on estimated respiratory
flow.
[0027] "Jamming", as described by Berthon-Jones, is the extent to
which the leak has not yet been compensated for, and usually
results from a rapid change in the leak. It is herein considered to
be a fuzzy logical quantity.
[0028] A leak conductance estimate G.sub.1 is calculated as
described above. Note that the time constant of the filters uses
preferably decreases as jamming increases, as described in
Berthon-Jones.
[0029] A second leak conductance estimate G.sub.2 is calculated
continuously, at the algorithmic sampling frequency or a lower
frequency (e.g. 10 Hz) which is still high compared with the
respiratory frequency. In a manner described below, the algorithm
identifies the position in the current breath, then attempts to
find time associated with the same position in the preceding
breath. If it fails to find such a position, it uses instead a time
10 seconds in the past. Between that time in the past and the
present, a window is established. Low-pass filtered mask flow and
low-pass filtered square root of mask pressure (filtered by a
non-breath-dependent method, such as a 1st-order LPF), typically
the low-pass filtered values used for the determination of G.sub.1,
are then further low-pass filtered by being averaged over this
window. The ratio of these window-averaged values is the leak
conductance estimate G.sub.2, which under conditions of stable leak
is extremely stable.
[0030] Because G.sub.2 responds rather slowly to changes in leak
conductance, it is inappropriate to use when the leak is changing
rapidly. Thus to the extent that there is jamming, G.sub.1 rather
than G.sub.2 is used. In the preferred implementation, if J is
jamming (a quantity in [0,1]), the conductance estimate G.sub.j is
used, given by
G.sub.j=JG.sub.1+(1-J)G.sub.2
Instantaneous leak is then straightforwardly calculated by
Q.sub.leak=G.sub.j {square root over (P.sub.mask)}.
[0031] Determining the Position of the Start of the Averaging
Window for G.sub.2
[0032] The aim is to determine the same position in the previous
breath as the patient, is at in the current breath. For this one
needs a concept of breath phase which is not just one of a small
set of categories, such as inspiration and expiration, but a
conceptually real-valued (in practice rational) variable which
increases progressively from the start of inspiration to the end of
expiration, potentially with a small finite number of jumps. Such a
concept is provided in Berthon-Jones Cheyne-Stokes patent,
WO98/012965, which is incorporated herein by reference. Breath
phase is there defined to be 0 at the start of inspiration, 0.5 at
the start of expiration, and approaches 1 at the end of expiration.
Given such a breath phase, one find the breath phase at the current
moment, then searches backward in time to find the same breath
phase in the previous breath. Because breath phase as estimated by
the system described by Berthon-Jones is not necessarily increasing
with time during a breath (neglecting the expiratory to inspiratory
transition, at which it must decrease) though typically it is
increasing with time during a breath, it is necessary to have an
algorithm which searches backward in time in such a way that a
point in the same breath with the same breath phase as the current
value is not identified as being in the previous breath. Such a
search algorithm is described below; this algorithm may fail under
exceptional circumstances, but is quite robust most of the time.
Because of jumps in phase, there may exist no point in the previous
breath with a phase equal to the phase associated with the current
moment, the latest time in the previous breath with a phase less
than or equal to the phase at the current moment is used instead.
On the other hand, a system which uses conventional flow thresholds
for triggering and cycling need not use a fuzzy logical system for
determining breath phase for the purpose of finding the same
position in the previous breath as in the current breath. Assuming
that during inspiration, the maximum time between the present until
the end of inspiration is known (typically determined at the start
of inspiration, but not necessarily), the breath phase at each
sampling interval is increased by such an amount that with equal
increments of that amount, the phase would reach 0.5 at the end of
inspiration. For example, in the simple case where a maximum
inspiratory time of 1.6 seconds was determined at the start of
inspiration, the phase would increase at a steady rate of 0.5/1.6
phase units/second. If cycling (transition to expiration) occurred
earlier, for example because respiratory flow fell below a cycling
threshold, the phase would at that point jump to 0.5. Similar
considerations apply during expiration, with rate of increase of
phase being the difference between 1 and the current phase divided
by the time remaining until the maximum expiratory time. If since
breath phase determined in this way is typically used only for the
purpose of determining the same position in the previous breath as
in the current breath, it is called "book-keeping" phase.
[0033] Regardless of the phase determination method used, whether
that of Berthon-Jones, "book-keeping" phase as described above, or
some other method, the search backward in time to find the latest
time in the preceding breath with a phase less than or equal to the
phase at the current moment is performed as follows (though it will
be appreciated that for "book-keeping" phase, simpler methods are
available).
[0034] Starting with the current phase, say .phi..sub.0, the
invention looks backwards in time for the most recent phase in the
interval [.phi..sub.0-0.75, .phi..sub.0-0.25]. The aim is to seek a
point in time at least 0.25 of a breath before the present. When
such a phase is found, the invention calculates
.phi..sub.1=.phi..sub.0-0.25 and looks backward for a phase in the
interval [.phi..sub.1-0.75, .phi..sub.1-0.25]. This is continued,
0.25 at a time, i.e. .phi..sub.i+1=.phi..sub.i-0.25. When a phase
is found which is in [.phi..sub.3-0.075, .phi..sub.3-0.25] the
iteration ceases, since this is just [.phi..sub.0-0.5,
.phi..sub.0]. If phase varied continuously this would have found
exactly .phi..sub.0; in reality it will most likely find
.phi..sub.0- ., where hopefully is small. By proceeding in this
manner we have some confidence that the phase has gone backward
rather than forward, since we have found phases in the 4 phase
quadrants before the present. This algorithm will regard two phase
transitions of 0.5 in succession as being movement backward, though
the actual direction is of course actually indeterminate. If this
algorithm fails to find a point between the present moment and a
time before the present which meets this criterion, we take the
start of the averaging window to be some reasonable maximum time
before the present, such as 10 seconds. As an implementation
detail, to reduce computational requirements, the leak, flow values
may be averaged over the last 0.1 seconds and stored in a buffer
accompanied by the associated breath phase, so that the search for
the last breath is performed in a buffer of 100 points, and done
every 0.1 seconds. The averaged leak estimate at the instantaneous
leak calculation frequency, e.g. 100 Hz, can then be calculated by
linear interpolation between the most recent averaged leak
conductance estimate and the averaged conductance leak estimate
just before it. In a servoventilator or other system using some
kind of measure of ventilation (such as half the absolute value of
respiratory flow, or a gross alveolar ventilation, or peak flow, or
some weighted average of flows determined during inspiration or
expiration) to adjust ventilatory support, when jamming is
observed, the system slams down or suspends changes in pressure
support. This is because respiratory flow estimates are not
reliable in the presence of jamming, and various measures of
ventilation based on respiratory flow are likely to overestimate
ventilation, leading for example in a servoventilator to
unwarranted withdrawal in ventilatory support because ventilation
appears to be above target ventilation. The extent of slowing down
of adjustment of respiratory support is preferably some increasing
function of the jamming. For example, if the calculated change in
respiratory support from that at the previous time that it was
calculated is some value .DELTA.S, then the adjusted change in
support would be k.DELTA.S, where for example k is 1 for
J.ltoreq.0.1, 0 for J.gtoreq.0.3, and taking linearly interpolated
values for intermediate values of J.
* * * * *